Wow, again, more real science on an Internet thread. Very nice.
+1
Rarely found, I know. For those getting bored please remind the vette owner's 'direct exhaust injection'?
To GoldNSXes points about variability and how testing over a longer period is valuable, and in theroy, this is quite true and is certainly fact. However, in a real world, many variable test, with very small change in test results, eliminating variables and/or magnifiying the effect of the product/item/change will help tell if there is a change. But, I'm done arguing that and I do agree with what you said.
I agree with you that the testing window has to be as small as possible to minimize disturbing error variance but not at the cost of only getting half or none of the effect. This would be like omissing variables and this is a big NONO in empirical statistic analysis. MvM's recalculation points to this problem. On the other hand we could minimize the test to a laboratory testing situation (isolating errorous variables) and this would be very interesting but it's simply too expensive AND may not be very realistic depending on how it is done.
In this test case, there are more variables in my opinion than there are in your opinion, from the sound of it. We had a fair degree of variance in our test results. I say fair degree, compared to how my car performs on a dyno. I get very similar runs (to 1 or 2 hp variance) per dyno run.
2 / 255 = .78% difference
In his tests, he had a .3 second difference on a test that took, at max, 5.88 seconds.
.3 / 5.88 = 5.1% difference
That's a big difference compare to more controlled situations. That is why I was stuck on the point to reduce test windows - to help eliminate this variance.
The std. dev. of 0.06 and 0.11 have been very small for an empirical test IMO. I agree with you that a dyno has less variance BUT I would not test the wheel weight effect with a dyno because the mass of the car is not moved and you only get the effect of the rear wheels. Our dynos here in Europe are far less quick than the ones in the US. One run takes about 40 sec. from 2-8k in 3rd gear and this is far less then if you accelerate the car on the road. So the acceleration of the wheel is very small and you get near to none effect. Actually the wheel acceleration is kept low as a dyno is intended to measure the engine output without being disturbed by unwanted variables (wheel weight in the case of a dyno).
Now, to MvM's point, you got your independent variables mixed up.
...
The wheels don't power the car. The engine does. The wheels must be accelerated, both linearly and rotationally, so their inertia resists this acceleration. But, the mass of the wheel does not provide a force for acceleration like the mass of a person does.
Fully agree.
This continues to be a great subject. I would still like to know how much of a difference lighter wheels make, however small.
Fully agree.
Let's do a summary of all the variables:
Dependent variable: acceleration time
Inpendent variables (only the ones that strongly influence the dependent variable or produce disturbing error variance). Which ones are most important?
- drivers technic (+-constant)
- car (was the same = variable held constant)
- vehicle mass (constant)
- tires (constant)
- tire pressure (I guess it was constant)
- fuel, was the same
- weather (temp. within 2-3 degrees, humidity, athmos. pressure, wind?)
- road (not constant! Of course it has been the same road but you don't drive the exactly same line or begin to accelerate at exactly the same point and hit every small bump at a different speed. Moreover error variance has been introduced by not driving on it always in the same direction. My test above should be a paired one, one for driving A-B and one for B-A (paired sample))
- wheel weight
Please feel free to add your independant var. on the list.
The AP-22 is an quite accurate meter. Of course you could do slightly better but at what costs?
From the list I guess there are only a few indep. var. that resulted in the already low error variance:
- temp. maybe but by how much?
- driver technic but by how much? I guess it's very low.
- ROAD! I think here lies the main cause of the (again even small) error variance.
Hypothesis:
The heavier wheels had a bigger variance than the lighter ones. Why? My hypothesis goes like this: Like I've argued in my 'audiophile' post above the heavier wheels have more problems to transmit the power to the (not perfectly flat) road and therefore are more prone to show one run a good time and the next run a bad time. You won't recognize this as wheel-spin or even reaction of the TCS as this happens only for a fraction of a second and by a small amount. It would have been interesting to see more insight into the higher std. deviation of the heavier wheels.
If this is the case we have measured two effects:
1. influence of lighter wheels on acceleration times on perfectly flat roads
+
2. influence of lighter wheels on the traction of the tires and therefore the influence on the acceleration times
In the test above we've measured both as a 'bundle'. We can't tell them from each other.
Or practically spoken: Do we have to? I'd say NO because we're interested in the acceleration times. That's why I'm not with isolated testing.